diff options
Diffstat (limited to 'gitlab/issues/target_arm/host_missing/accel_missing/1627.toml')
| -rw-r--r-- | gitlab/issues/target_arm/host_missing/accel_missing/1627.toml | 49 |
1 files changed, 49 insertions, 0 deletions
diff --git a/gitlab/issues/target_arm/host_missing/accel_missing/1627.toml b/gitlab/issues/target_arm/host_missing/accel_missing/1627.toml new file mode 100644 index 00000000..7064f8d2 --- /dev/null +++ b/gitlab/issues/target_arm/host_missing/accel_missing/1627.toml @@ -0,0 +1,49 @@ +id = 1627 +title = "Aarch64: VTCR.T0SZ / iasize test for Aarch32 guests wrong" +state = "closed" +created_at = "2023-04-27T15:23:21.691Z" +closed_at = "2023-05-13T08:35:45.595Z" +labels = ["target: arm"] +url = "https://gitlab.com/qemu-project/qemu/-/issues/1627" +host-os = "Debian/Sid/Bookworm" +host-arch = "x86 / 64-bit" +qemu-version = "v8.0.0-248-g732d6603d0-dirty" +guest-os = "L4Re (custom)" +guest-arch = "ARM (Aarch64)" +description = """With QEMU 8 we are no longer able to execute Aarch32 guest code on an Aarch64 host. We use virtualization for the QEMU guest: +- The QEMU guest kernel (L4Re kernel) runs at EL2 in AArch64 mode. +- The L4Re guest code runs at EL1 in AAarch32 mode. + +It seems that the check for T0SZ / iasize in `ptw.c` / `check_s2_mmu_setup()` is too strict: +``` +if (is_aa64) { + /* + * AArch64.S2InvalidTxSZ: While we checked tsz_oob near the top of + * get_phys_addr_lpae, that used aa64_va_parameters which apply + * to aarch64. If Stage1 is aarch32, the min_txsz is larger. + * See AArch64.S2MinTxSZ, where min_tsz is 24, translated to + * inputsize is 64 - 24 = 40. + */ + if (iasize < 40 && !arm_el_is_aa64(&cpu->env, 1)) { + goto fail; + } +``` +The above test fails for us when executing Aarch32 EL1 code on Aarch64 EL2. + +Please note that the comment talks about `S2MinTxSZ` / `min_tsz`, so if the **minimum** value of `T0SZ` is 24, then the **maximum** value of `iasize` is `64-24=40` so the following comparison would be more appropriate (I replaces `<` by `>`): +``` +if (iasize > 40 && !arm_el_is_aa64(&cpu->env, 1)) { + goto fail; +} +``` +However, the minimum value of `VTCR_EL2.T0SZ` is either 16 or 12, see `VTCR_EL2.DS`: +- `VTCR_EL2.DS=0b0`: **minimum** value of `VTCR_EL2.T0SZ` is 16 => **maximum** value of `iasize` is 48, +- `VTCR_EL2.DS=0b1`: **minimum** value of `VTCR_EL2.T0SZ` is 12 => **maximum** value of `iasize` is 52. + +Regarding the minimum of `iasize` / maximum of `VTCR_EL2.T0SZ`, see `ID_AA64MMFR_EL1.ST`: +- `ID_AA64MMFR2_EL1.ST=0b0000`: **maximum** value of `VTCR_EL2.T0SZ` is 39 => **minimum** value of `iasize` is 25, +- `ID_AA64MMFR2_EL1.ST=0b0001`: **maximum** value of `VTCR_EL2.T0SZ` is 48 => **minimum** value of `iasize` is 16 (or 47/17 for 64KiB granules). + +Our system executes Aarch32 EL1 code fine on Aarch64 EL2 if I weaken the comparison.""" +reproduce = "n/a" +additional = """Sorry for not providing a test build but I'm not sure if it's worth to provide a custom build of our L4Re system, but I will happily provide one if you insist.""" |